1Password Hands Claude Your Saved Passwords

1Password shipped a browser integration that lets Anthropic's Claude use saved usernames and passwords directly, per The Verge. Any page the agent browses becomes a possible route to steal live credentials, not a workflow annoyance you can shrug off.

The Verge covers 1Password as the first password manager to sanction agent access to a vault instead of treating that access as an attack. Prompt injection stopped being about tricking a chatbot into saying something dumb. Now it's a path to your bank login.

Autofill for agents is the obvious next step, and the reasoning behind the integration makes sense on its face. But the threat model changed under it, and the mitigations haven't caught up.

A prompt-injected page doesn't need to fool a person anymore. It just needs to fool Claude.

The Verge's report doesn't detail how the integration handles a page that tries to trick Claude into handing off a credential mid-session. That gap is the whole story.